-
Notifications
You must be signed in to change notification settings - Fork 22
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Poor performance on custom instance #11
Comments
Sorry for the late response. Could you provide more details? For example, what diffusers version did you use? Do you mean you can not generate good results with your concept model and your source image, or your concept model and the provided Prof. Feifei Li image? Also, one tip is to use around 20 images to train the DreamBooth, which might increase the performance. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Thank you for your work, the performance of the provided demo is excellent.
I follow your instruction and train my own concept model by DreamBooth with text-encoder fine-tuned.
Through validation, it can generate some good imgs itself.
However, when I use it as the concept model and run the swap func, I cannot get meaningful results, even in any combinations of [cross_map_replace_steps, self_output_replace_steps, self_map_replace_steps].
I wonder which part is wrong?
I think the step1 may have some mistakes, because when I change it to the provided taylor, the results are good with the same source img, but the concept model I trained can already generate the target concept.
The text was updated successfully, but these errors were encountered: